/* $OpenBSD: nfs_socket.c,v 1.56 2007/12/28 17:25:18 thib Exp $ */ /* $NetBSD: nfs_socket.c,v 1.27 1996/04/15 20:20:00 thorpej Exp $ */ /* * Copyright (c) 1989, 1991, 1993, 1995 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 */ /* * Socket operations for use by nfs */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define TRUE 1 #define FALSE 0 /* * Estimate rto for an nfs rpc sent via. an unreliable datagram. * Use the mean and mean deviation of rtt for the appropriate type of rpc * for the frequent rpcs and a default for the others. * The justification for doing "other" this way is that these rpcs * happen so infrequently that timer est. would probably be stale. * Also, since many of these rpcs are * non-idempotent, a conservative timeout is desired. * getattr, lookup - A+2D * read, write - A+4D * other - nm_timeo */ #define NFS_RTO(n, t) \ ((t) == 0 ? (n)->nm_timeo : \ ((t) < 3 ? \ (((((n)->nm_srtt[t-1] + 3) >> 2) + (n)->nm_sdrtt[t-1] + 1) >> 1) : \ ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] /* * External data, mostly RPC constants in XDR form */ extern u_int32_t rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, rpc_auth_unix, rpc_msgaccepted, rpc_call, rpc_autherr, rpc_auth_kerb; extern u_int32_t nfs_prog; extern struct nfsstats nfsstats; extern int nfsv3_procid[NFS_NPROCS]; extern int nfs_ticks; /* * Defines which timer to use for the procnum. * 0 - default * 1 - getattr * 2 - lookup * 3 - read * 4 - write */ static int proct[NFS_NPROCS] = { 0, 1, 0, 2, 1, 3, 3, 4, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 0, 0, 0, 0, 0, 0, }; /* * There is a congestion window for outstanding rpcs maintained per mount * point. The cwnd size is adjusted in roughly the way that: * Van Jacobson, Congestion avoidance and Control, In "Proceedings of * SIGCOMM '88". ACM, August 1988. * describes for TCP. The cwnd size is chopped in half on a retransmit timeout * and incremented by 1/cwnd when each rpc reply is received and a full cwnd * of rpcs is in progress. * (The sent count and cwnd are scaled for integer arith.) * Variants of "slow start" were tried and were found to be too much of a * performance hit (ave. rtt 3 times larger), * I suspect due to the large rtt that nfs rpcs have. */ #define NFS_CWNDSCALE 256 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; int nfsrtton = 0; struct nfsrtt nfsrtt; void nfs_realign(struct mbuf **, int); void nfs_realign_fixup(struct mbuf *, struct mbuf *, unsigned int *); unsigned int nfs_realign_test = 0; unsigned int nfs_realign_count = 0; struct nfsreqhead nfs_reqq; /* * Initialize sockets and congestion for a new NFS connection. * We do not free the sockaddr if error. */ int nfs_connect(nmp, rep) struct nfsmount *nmp; struct nfsreq *rep; { struct socket *so; int s, error, rcvreserve, sndreserve; struct sockaddr *saddr; struct sockaddr_in *sin; struct mbuf *m; nmp->nm_so = (struct socket *)0; saddr = mtod(nmp->nm_nam, struct sockaddr *); error = socreate(saddr->sa_family, &nmp->nm_so, nmp->nm_sotype, nmp->nm_soproto); if (error) goto bad; so = nmp->nm_so; nmp->nm_soflags = so->so_proto->pr_flags; /* * Some servers require that the client port be a reserved port number. * We always allocate a reserved port, as this prevents filehandle * disclosure through UDP port capture. */ if (saddr->sa_family == AF_INET) { struct mbuf *mopt; int *ip; MGET(mopt, M_WAIT, MT_SOOPTS); mopt->m_len = sizeof(int); ip = mtod(mopt, int *); *ip = IP_PORTRANGE_LOW; error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt); if (error) goto bad; MGET(m, M_WAIT, MT_SONAME); sin = mtod(m, struct sockaddr_in *); sin->sin_len = m->m_len = sizeof (struct sockaddr_in); sin->sin_family = AF_INET; sin->sin_addr.s_addr = INADDR_ANY; sin->sin_port = htons(0); error = sobind(so, m); m_freem(m); if (error) goto bad; MGET(mopt, M_WAIT, MT_SOOPTS); mopt->m_len = sizeof(int); ip = mtod(mopt, int *); *ip = IP_PORTRANGE_DEFAULT; error = sosetopt(so, IPPROTO_IP, IP_PORTRANGE, mopt); if (error) goto bad; } /* * Protocols that do not require connections may be optionally left * unconnected for servers that reply from a port other than NFS_PORT. */ if (nmp->nm_flag & NFSMNT_NOCONN) { if (nmp->nm_soflags & PR_CONNREQUIRED) { error = ENOTCONN; goto bad; } } else { error = soconnect(so, nmp->nm_nam); if (error) goto bad; /* * Wait for the connection to complete. Cribbed from the * connect system call but with the wait timing out so * that interruptible mounts don't hang here for a long time. */ s = splsoftnet(); while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { (void) tsleep((caddr_t)&so->so_timeo, PSOCK, "nfscon", 2 * hz); if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0 && rep && (error = nfs_sigintr(nmp, rep, rep->r_procp)) != 0){ so->so_state &= ~SS_ISCONNECTING; splx(s); goto bad; } } if (so->so_error) { error = so->so_error; so->so_error = 0; splx(s); goto bad; } splx(s); } /* * Always set receive timeout to detect server crash and reconnect. * Otherwise, we can get stuck in soreceive forever. */ so->so_rcv.sb_timeo = (5 * hz); if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) so->so_snd.sb_timeo = (5 * hz); else so->so_snd.sb_timeo = 0; if (nmp->nm_sotype == SOCK_DGRAM) { sndreserve = nmp->nm_wsize + NFS_MAXPKTHDR; rcvreserve = max(nmp->nm_rsize, nmp->nm_readdirsize) + NFS_MAXPKTHDR; } else if (nmp->nm_sotype == SOCK_SEQPACKET) { sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + NFS_MAXPKTHDR) * 2; } else { if (nmp->nm_sotype != SOCK_STREAM) panic("nfscon sotype"); if (so->so_proto->pr_flags & PR_CONNREQUIRED) { MGET(m, M_WAIT, MT_SOOPTS); *mtod(m, int32_t *) = 1; m->m_len = sizeof(int32_t); sosetopt(so, SOL_SOCKET, SO_KEEPALIVE, m); } if (so->so_proto->pr_protocol == IPPROTO_TCP) { MGET(m, M_WAIT, MT_SOOPTS); *mtod(m, int32_t *) = 1; m->m_len = sizeof(int32_t); sosetopt(so, IPPROTO_TCP, TCP_NODELAY, m); } sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + sizeof (u_int32_t)) * 2; rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + sizeof (u_int32_t)) * 2; } error = soreserve(so, sndreserve, rcvreserve); if (error) goto bad; so->so_rcv.sb_flags |= SB_NOINTR; so->so_snd.sb_flags |= SB_NOINTR; /* Initialize other non-zero congestion variables */ nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = (NFS_TIMEO << 3); nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = nmp->nm_sdrtt[3] = 0; nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ nmp->nm_sent = 0; nmp->nm_timeouts = 0; return (0); bad: nfs_disconnect(nmp); return (error); } /* * Reconnect routine: * Called when a connection is broken on a reliable protocol. * - clean up the old socket * - nfs_connect() again * - set R_MUSTRESEND for all outstanding requests on mount point * If this fails the mount point is DEAD! * nb: Must be called with the nfs_sndlock() set on the mount point. */ int nfs_reconnect(rep) struct nfsreq *rep; { struct nfsreq *rp; struct nfsmount *nmp = rep->r_nmp; int error; nfs_disconnect(nmp); while ((error = nfs_connect(nmp, rep)) != 0) { if (error == EINTR || error == ERESTART) return (EINTR); (void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0); } /* * Loop through outstanding request list and fix up all requests * on old socket. */ TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { if (rp->r_nmp == nmp) { rp->r_flags |= R_MUSTRESEND; rp->r_rexmit = 0; } } return (0); } /* * NFS disconnect. Clean up and unlink. */ void nfs_disconnect(nmp) struct nfsmount *nmp; { struct socket *so; if (nmp->nm_so) { so = nmp->nm_so; nmp->nm_so = (struct socket *)0; soshutdown(so, SHUT_RDWR); soclose(so); } } /* * This is the nfs send routine. For connection based socket types, it * must be called with an nfs_sndlock() on the socket. * "rep == NULL" indicates that it has been called from a server. * For the client side: * - return EINTR if the RPC is terminated, 0 otherwise * - set R_MUSTRESEND if the send fails for any reason * - do any cleanup required by recoverable socket errors (???) * For the server side: * - return EINTR or ERESTART if interrupted by a signal * - return EPIPE if a connection is lost for connection based sockets (TCP...) * - do any cleanup required by recoverable socket errors (???) */ int nfs_send(so, nam, top, rep) struct socket *so; struct mbuf *nam; struct mbuf *top; struct nfsreq *rep; { struct mbuf *sendnam; int error, soflags, flags; if (rep) { if (rep->r_flags & R_SOFTTERM) { m_freem(top); return (EINTR); } if ((so = rep->r_nmp->nm_so) == NULL) { rep->r_flags |= R_MUSTRESEND; m_freem(top); return (0); } rep->r_flags &= ~R_MUSTRESEND; soflags = rep->r_nmp->nm_soflags; } else soflags = so->so_proto->pr_flags; if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED)) sendnam = (struct mbuf *)0; else sendnam = nam; if (so->so_type == SOCK_SEQPACKET) flags = MSG_EOR; else flags = 0; error = sosend(so, sendnam, (struct uio *)0, top, (struct mbuf *)0, flags); if (error) { if (rep) { /* * Deal with errors for the client side. */ if (rep->r_flags & R_SOFTTERM) error = EINTR; else rep->r_flags |= R_MUSTRESEND; } /* * Handle any recoverable (soft) socket errors here. (???) */ if (error != EINTR && error != ERESTART && error != EWOULDBLOCK && error != EPIPE) error = 0; } return (error); } #ifdef NFSCLIENT /* * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all * done by soreceive(), but for SOCK_STREAM we must deal with the Record * Mark and consolidate the data into a new mbuf list. * nb: Sometimes TCP passes the data up to soreceive() in long lists of * small mbufs. * For SOCK_STREAM we must be very careful to read an entire record once * we have read any of it, even if the system call has been interrupted. */ int nfs_receive(rep, aname, mp) struct nfsreq *rep; struct mbuf **aname; struct mbuf **mp; { struct socket *so; struct uio auio; struct iovec aio; struct mbuf *m; struct mbuf *control; u_int32_t len; struct mbuf **getnam; int error, sotype, rcvflg; struct proc *p = curproc; /* XXX */ /* * Set up arguments for soreceive() */ *mp = (struct mbuf *)0; *aname = (struct mbuf *)0; sotype = rep->r_nmp->nm_sotype; /* * For reliable protocols, lock against other senders/receivers * in case a reconnect is necessary. * For SOCK_STREAM, first get the Record Mark to find out how much * more there is to get. * We must lock the socket against other receivers * until we have an entire rpc request/reply. */ if (sotype != SOCK_DGRAM) { error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); if (error) return (error); tryagain: /* * Check for fatal errors and resending request. */ /* * Ugh: If a reconnect attempt just happened, nm_so * would have changed. NULL indicates a failed * attempt that has essentially shut down this * mount point. */ if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) { nfs_sndunlock(&rep->r_nmp->nm_flag); return (EINTR); } so = rep->r_nmp->nm_so; if (!so) { error = nfs_reconnect(rep); if (error) { nfs_sndunlock(&rep->r_nmp->nm_flag); return (error); } goto tryagain; } while (rep->r_flags & R_MUSTRESEND) { m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT); nfsstats.rpcretries++; rep->r_rtt = 0; rep->r_flags &= ~R_TIMING; error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); if (error) { if (error == EINTR || error == ERESTART || (error = nfs_reconnect(rep)) != 0) { nfs_sndunlock(&rep->r_nmp->nm_flag); return (error); } goto tryagain; } } nfs_sndunlock(&rep->r_nmp->nm_flag); if (sotype == SOCK_STREAM) { aio.iov_base = (caddr_t) &len; aio.iov_len = sizeof(u_int32_t); auio.uio_iov = &aio; auio.uio_iovcnt = 1; auio.uio_segflg = UIO_SYSSPACE; auio.uio_rw = UIO_READ; auio.uio_offset = 0; auio.uio_resid = sizeof(u_int32_t); auio.uio_procp = p; do { rcvflg = MSG_WAITALL; error = soreceive(so, (struct mbuf **)0, &auio, (struct mbuf **)0, (struct mbuf **)0, &rcvflg); if (error == EWOULDBLOCK && rep) { if (rep->r_flags & R_SOFTTERM) return (EINTR); /* * looks like the server died after it * received the request, make sure * that we will retransmit and we * don't get stuck here forever. */ if (rep->r_rexmit >= rep->r_nmp->nm_retry) { nfsstats.rpctimeouts++; error = EPIPE; } } } while (error == EWOULDBLOCK); if (!error && auio.uio_resid > 0) { log(LOG_INFO, "short receive (%d/%d) from nfs server %s\n", sizeof(u_int32_t) - auio.uio_resid, sizeof(u_int32_t), rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); error = EPIPE; } if (error) goto errout; len = ntohl(len) & ~0x80000000; /* * This is SERIOUS! We are out of sync with the sender * and forcing a disconnect/reconnect is all I can do. */ if (len > NFS_MAXPACKET) { log(LOG_ERR, "%s (%d) from nfs server %s\n", "impossible packet length", len, rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); error = EFBIG; goto errout; } auio.uio_resid = len; do { rcvflg = MSG_WAITALL; error = soreceive(so, (struct mbuf **)0, &auio, mp, (struct mbuf **)0, &rcvflg); } while (error == EWOULDBLOCK || error == EINTR || error == ERESTART); if (!error && auio.uio_resid > 0) { log(LOG_INFO, "short receive (%d/%d) from nfs server %s\n", len - auio.uio_resid, len, rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); error = EPIPE; } } else { /* * NB: Since uio_resid is big, MSG_WAITALL is ignored * and soreceive() will return when it has either a * control msg or a data msg. * We have no use for control msg., but must grab them * and then throw them away so we know what is going * on. */ auio.uio_resid = len = 100000000; /* Anything Big */ auio.uio_procp = p; do { rcvflg = 0; error = soreceive(so, (struct mbuf **)0, &auio, mp, &control, &rcvflg); if (control) m_freem(control); if (error == EWOULDBLOCK && rep) { if (rep->r_flags & R_SOFTTERM) return (EINTR); } } while (error == EWOULDBLOCK || (!error && *mp == NULL && control)); if ((rcvflg & MSG_EOR) == 0) printf("Egad!!\n"); if (!error && *mp == NULL) error = EPIPE; len -= auio.uio_resid; } errout: if (error && error != EINTR && error != ERESTART) { m_freem(*mp); *mp = (struct mbuf *)0; if (error != EPIPE) log(LOG_INFO, "receive error %d from nfs server %s\n", error, rep->r_nmp->nm_mountp->mnt_stat.f_mntfromname); error = nfs_sndlock(&rep->r_nmp->nm_flag, rep); if (!error) { error = nfs_reconnect(rep); if (!error) goto tryagain; nfs_sndunlock(&rep->r_nmp->nm_flag); } } } else { if ((so = rep->r_nmp->nm_so) == NULL) return (EACCES); if (so->so_state & SS_ISCONNECTED) getnam = (struct mbuf **)0; else getnam = aname; auio.uio_resid = len = 1000000; auio.uio_procp = p; do { rcvflg = 0; error = soreceive(so, getnam, &auio, mp, (struct mbuf **)0, &rcvflg); if (error == EWOULDBLOCK && (rep->r_flags & R_SOFTTERM)) return (EINTR); } while (error == EWOULDBLOCK); len -= auio.uio_resid; } if (error) { m_freem(*mp); *mp = (struct mbuf *)0; } /* * Search for any mbufs that are not a multiple of 4 bytes long * or with m_data not longword aligned. * These could cause pointer alignment problems, so copy them to * well aligned mbufs. */ nfs_realign(mp, 5 * NFSX_UNSIGNED); return (error); } /* * Implement receipt of reply on a socket. * We must search through the list of received datagrams matching them * with outstanding requests using the xid, until ours is found. */ /* ARGSUSED */ int nfs_reply(myrep) struct nfsreq *myrep; { struct nfsreq *rep; struct nfsmount *nmp = myrep->r_nmp; int32_t t1; struct mbuf *mrep, *nam, *md; u_int32_t rxid, *tl; caddr_t dpos, cp2; int error; /* * Loop around until we get our own reply */ for (;;) { /* * Lock against other receivers so that I don't get stuck in * sbwait() after someone else has received my reply for me. * Also necessary for connection based protocols to avoid * race conditions during a reconnect. */ error = nfs_rcvlock(myrep); if (error) return (error); /* Already received, bye bye */ if (myrep->r_mrep != NULL) { nfs_rcvunlock(&nmp->nm_flag); return (0); } /* * Get the next Rpc reply off the socket */ error = nfs_receive(myrep, &nam, &mrep); nfs_rcvunlock(&nmp->nm_flag); if (error) { /* * Ignore routing errors on connectionless protocols?? */ if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) { if (nmp->nm_so) nmp->nm_so->so_error = 0; if (myrep->r_flags & R_GETONEREP) return (0); continue; } return (error); } if (nam) m_freem(nam); /* * Get the xid and check that it is an rpc reply */ md = mrep; dpos = mtod(md, caddr_t); nfsm_dissect(tl, u_int32_t *, 2*NFSX_UNSIGNED); rxid = *tl++; if (*tl != rpc_reply) { nfsstats.rpcinvalid++; m_freem(mrep); nfsmout: if (myrep->r_flags & R_GETONEREP) return (0); continue; } /* * Loop through the request list to match up the reply * Iff no match, just drop the datagram */ for (rep = TAILQ_FIRST(&nfs_reqq); rep != NULL; rep = TAILQ_NEXT(rep, r_chain)) { if (rep->r_mrep == NULL && rxid == rep->r_xid) { /* Found it.. */ rep->r_mrep = mrep; rep->r_md = md; rep->r_dpos = dpos; if (nfsrtton) { struct rttl *rt; rt = &nfsrtt.rttl[nfsrtt.pos]; rt->proc = rep->r_procnum; rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); rt->sent = nmp->nm_sent; rt->cwnd = nmp->nm_cwnd; rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; rt->fsid = nmp->nm_mountp->mnt_stat.f_fsid; getmicrotime(&rt->tstamp); if (rep->r_flags & R_TIMING) rt->rtt = rep->r_rtt; else rt->rtt = 1000000; nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; } /* * Update congestion window. * Do the additive increase of * one rpc/rtt. */ if (nmp->nm_cwnd <= nmp->nm_sent) { nmp->nm_cwnd += (NFS_CWNDSCALE * NFS_CWNDSCALE + (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; if (nmp->nm_cwnd > NFS_MAXCWND) nmp->nm_cwnd = NFS_MAXCWND; } rep->r_flags &= ~R_SENT; nmp->nm_sent -= NFS_CWNDSCALE; /* * Update rtt using a gain of 0.125 on the mean * and a gain of 0.25 on the deviation. */ if (rep->r_flags & R_TIMING) { /* * Since the timer resolution of * NFS_HZ is so course, it can often * result in r_rtt == 0. Since * r_rtt == N means that the actual * rtt is between N+dt and N+2-dt ticks, * add 1. */ t1 = rep->r_rtt + 1; t1 -= (NFS_SRTT(rep) >> 3); NFS_SRTT(rep) += t1; if (t1 < 0) t1 = -t1; t1 -= (NFS_SDRTT(rep) >> 2); NFS_SDRTT(rep) += t1; } nmp->nm_timeouts = 0; break; } } /* * If not matched to a request, drop it. * If it's mine, get out. */ if (rep == 0) { nfsstats.rpcunexpected++; m_freem(mrep); } else if (rep == myrep) { if (rep->r_mrep == NULL) panic("nfsreply nil"); return (0); } if (myrep->r_flags & R_GETONEREP) return (0); } } /* * nfs_request - goes something like this * - fill in request struct * - links it into list * - calls nfs_send() for first transmit * - calls nfs_receive() to get reply * - break down rpc header and return with nfs reply pointed to * by mrep or error * nb: always frees up mreq mbuf list */ int nfs_request(vp, mrest, procnum, procp, cred, mrp, mdp, dposp) struct vnode *vp; struct mbuf *mrest; int procnum; struct proc *procp; struct ucred *cred; struct mbuf **mrp; struct mbuf **mdp; caddr_t *dposp; { struct mbuf *m, *mrep; struct nfsreq *rep; u_int32_t *tl; int i; struct nfsmount *nmp; struct mbuf *md; time_t waituntil; caddr_t dpos, cp2; int t1, s, error = 0, mrest_len; int trylater_delay; trylater_delay = NFS_MINTIMEO; nmp = VFSTONFS(vp->v_mount); rep = pool_get(&nfsreqpl, PR_WAITOK); rep->r_nmp = nmp; rep->r_vp = vp; rep->r_procp = procp; rep->r_procnum = procnum; i = 0; m = mrest; while (m) { i += m->m_len; m = m->m_next; } mrest_len = i; /* Get the RPC header with authorization. */ nfsm_rpchead(rep, cred, RPCAUTH_UNIX, mrest, mrest_len); m = rep->r_mreq; /* * For stream protocols, insert a Sun RPC Record Mark. */ if (nmp->nm_sotype == SOCK_STREAM) { M_PREPEND(m, NFSX_UNSIGNED, M_WAIT); *mtod(m, u_int32_t *) = htonl(0x80000000 | (m->m_pkthdr.len - NFSX_UNSIGNED)); } tryagain: if (nmp->nm_flag & NFSMNT_SOFT) rep->r_retry = nmp->nm_retry; else rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ rep->r_rtt = rep->r_rexmit = 0; if (proct[procnum] > 0) rep->r_flags = R_TIMING; else rep->r_flags = 0; rep->r_mrep = NULL; /* * Do the client side RPC. */ nfsstats.rpcrequests++; /* * Chain request into list of outstanding requests. Be sure * to put it LAST so timer finds oldest requests first. */ s = splsoftnet(); TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); /* * If backing off another request or avoiding congestion, don't * send this one now but let timer do it. If not timing a request, * do it now. */ if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || (nmp->nm_flag & NFSMNT_DUMBTIMR) || nmp->nm_sent < nmp->nm_cwnd)) { splx(s); if (nmp->nm_soflags & PR_CONNREQUIRED) error = nfs_sndlock(&nmp->nm_flag, rep); if (!error) { error = nfs_send(nmp->nm_so, nmp->nm_nam, m_copym(m, 0, M_COPYALL, M_WAIT), rep); if (nmp->nm_soflags & PR_CONNREQUIRED) nfs_sndunlock(&nmp->nm_flag); } if (!error && (rep->r_flags & R_MUSTRESEND) == 0) { nmp->nm_sent += NFS_CWNDSCALE; rep->r_flags |= R_SENT; } } else { splx(s); rep->r_rtt = -1; } /* * Wait for the reply from our send or the timer's. */ if (!error || error == EPIPE) error = nfs_reply(rep); /* * RPC done, unlink the request. */ s = splsoftnet(); TAILQ_REMOVE(&nfs_reqq, rep, r_chain); splx(s); /* * Decrement the outstanding request count. */ if (rep->r_flags & R_SENT) { rep->r_flags &= ~R_SENT; /* paranoia */ nmp->nm_sent -= NFS_CWNDSCALE; } /* * If there was a successful reply and a tprintf msg. * tprintf a response. */ if (!error && (rep->r_flags & R_TPRINTFMSG)) nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, "is alive again"); mrep = rep->r_mrep; md = rep->r_md; dpos = rep->r_dpos; if (error) { m_freem(rep->r_mreq); pool_put(&nfsreqpl, rep); return (error); } /* * break down the rpc header and check if ok */ nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); if (*tl++ == rpc_msgdenied) { if (*tl == rpc_mismatch) error = EOPNOTSUPP; else error = EACCES; m_freem(mrep); m_freem(rep->r_mreq); pool_put(&nfsreqpl, rep); return (error); } /* * Since we only support RPCAUTH_UNIX atm we step over the * reply verifer type, and if the (error) case that there really * is any data init, we advance over it. */ tl++; /* Step over verifer type */ i = fxdr_unsigned(int32_t, *tl); if (i > 0) nfsm_adv(nfsm_rndup(i)); /* Should not happen */ nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); /* 0 == ok */ if (*tl == 0) { nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl != 0) { error = fxdr_unsigned(int, *tl); if ((nmp->nm_flag & NFSMNT_NFSV3) && error == NFSERR_TRYLATER) { m_freem(mrep); error = 0; waituntil = time_second + trylater_delay; while (time_second < waituntil) (void) tsleep((caddr_t)&lbolt, PSOCK, "nqnfstry", 0); trylater_delay *= NFS_TIMEOUTMUL; if (trylater_delay > NFS_MAXTIMEO) trylater_delay = NFS_MAXTIMEO; goto tryagain; } /* * If the File Handle was stale, invalidate the * lookup cache, just in case. */ if (error == ESTALE) cache_purge(vp); if (nmp->nm_flag & NFSMNT_NFSV3 || error == ESTALE) { *mrp = mrep; *mdp = md; *dposp = dpos; error |= NFSERR_RETERR; } else m_freem(mrep); m_freem(rep->r_mreq); pool_put(&nfsreqpl, rep); return (error); } *mrp = mrep; *mdp = md; *dposp = dpos; m_freem(rep->r_mreq); pool_put(&nfsreqpl, rep); return (0); } m_freem(mrep); error = EPROTONOSUPPORT; nfsmout: m_freem(rep->r_mreq); pool_put(&nfsreqpl, rep); return (error); } #endif /* NFSCLIENT */ /* * Generate the rpc reply header * siz arg. is used to decide if adding a cluster is worthwhile */ int nfs_rephead(siz, nd, slp, err, mrq, mbp, bposp) int siz; struct nfsrv_descript *nd; struct nfssvc_sock *slp; int err; struct mbuf **mrq; struct mbuf **mbp; caddr_t *bposp; { u_int32_t *tl; struct mbuf *mreq; caddr_t bpos; struct mbuf *mb, *mb2; MGETHDR(mreq, M_WAIT, MT_DATA); mb = mreq; /* * If this is a big reply, use a cluster else * try and leave leading space for the lower level headers. */ siz += RPC_REPLYSIZ; if (siz >= max_datalen) { MCLGET(mreq, M_WAIT); } else mreq->m_data += max_hdr; tl = mtod(mreq, u_int32_t *); mreq->m_len = 6 * NFSX_UNSIGNED; bpos = ((caddr_t)tl) + mreq->m_len; *tl++ = txdr_unsigned(nd->nd_retxid); *tl++ = rpc_reply; if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { *tl++ = rpc_msgdenied; if (err & NFSERR_AUTHERR) { *tl++ = rpc_autherr; *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); mreq->m_len -= NFSX_UNSIGNED; bpos -= NFSX_UNSIGNED; } else { *tl++ = rpc_mismatch; *tl++ = txdr_unsigned(RPC_VER2); *tl = txdr_unsigned(RPC_VER2); } } else { *tl++ = rpc_msgaccepted; /* * For Kerberos authentication, we must send the nickname * verifier back, otherwise just RPCAUTH_NULL. */ if (nd->nd_flag & ND_KERBFULL) { struct nfsuid *nuidp; struct timeval ktvin, ktvout; LIST_FOREACH(nuidp, NUIDHASH(slp, nd->nd_cr.cr_uid), nu_hash) { if (nuidp->nu_cr.cr_uid == nd->nd_cr.cr_uid && (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), &nuidp->nu_haddr, nd->nd_nam2))) break; } if (nuidp) { ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec); *tl++ = rpc_auth_kerb; *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); *tl = ktvout.tv_sec; nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED); *tl++ = ktvout.tv_usec; *tl++ = txdr_unsigned(nuidp->nu_cr.cr_uid); } else { *tl++ = 0; *tl++ = 0; } } else { *tl++ = 0; *tl++ = 0; } switch (err) { case EPROGUNAVAIL: *tl = txdr_unsigned(RPC_PROGUNAVAIL); break; case EPROGMISMATCH: *tl = txdr_unsigned(RPC_PROGMISMATCH); nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(2); *tl = txdr_unsigned(3); break; case EPROCUNAVAIL: *tl = txdr_unsigned(RPC_PROCUNAVAIL); break; case EBADRPC: *tl = txdr_unsigned(RPC_GARBAGE); break; default: *tl = 0; if (err != NFSERR_RETVOID) { nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); if (err) *tl = txdr_unsigned(nfsrv_errmap(nd, err)); else *tl = 0; } break; }; } *mrq = mreq; if (mbp != NULL) *mbp = mb; *bposp = bpos; if (err != 0 && err != NFSERR_RETVOID) nfsstats.srvrpc_errs++; return (0); } /* * Nfs timer routine * Scan the nfsreq list and retranmit any requests that have timed out * To avoid retransmission attempts on STREAM sockets (in the future) make * sure to set the r_retry field to 0 (implies nm_retry == 0). */ void nfs_timer(arg) void *arg; { struct timeout *to = (struct timeout *)arg; struct nfsreq *rep; struct mbuf *m; struct socket *so; struct nfsmount *nmp; int timeo; int s, error; #ifdef NFSSERVER struct nfssvc_sock *slp; struct timeval tv; u_quad_t cur_usec; #endif s = splsoftnet(); TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { nmp = rep->r_nmp; if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) continue; if (nfs_sigintr(nmp, rep, rep->r_procp)) { rep->r_flags |= R_SOFTTERM; continue; } if (rep->r_rtt >= 0) { rep->r_rtt++; if (nmp->nm_flag & NFSMNT_DUMBTIMR) timeo = nmp->nm_timeo; else timeo = NFS_RTO(nmp, proct[rep->r_procnum]); if (nmp->nm_timeouts > 0) timeo *= nfs_backoff[nmp->nm_timeouts - 1]; if (rep->r_rtt <= timeo) continue; if (nmp->nm_timeouts < 8) nmp->nm_timeouts++; } /* * Check for server not responding */ if ((rep->r_flags & R_TPRINTFMSG) == 0 && rep->r_rexmit > nmp->nm_deadthresh) { nfs_msg(rep->r_procp, nmp->nm_mountp->mnt_stat.f_mntfromname, "not responding"); rep->r_flags |= R_TPRINTFMSG; } if (rep->r_rexmit >= rep->r_retry) { /* too many */ nfsstats.rpctimeouts++; rep->r_flags |= R_SOFTTERM; continue; } if (nmp->nm_sotype != SOCK_DGRAM) { if (++rep->r_rexmit > NFS_MAXREXMIT) rep->r_rexmit = NFS_MAXREXMIT; continue; } if ((so = nmp->nm_so) == NULL) continue; /* * If there is enough space and the window allows.. * Resend it * Set r_rtt to -1 in case we fail to send it now. */ rep->r_rtt = -1; if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len && ((nmp->nm_flag & NFSMNT_DUMBTIMR) || (rep->r_flags & R_SENT) || nmp->nm_sent < nmp->nm_cwnd) && (m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))){ if ((nmp->nm_flag & NFSMNT_NOCONN) == 0) error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, (struct mbuf *)0, (struct mbuf *)0); else error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, m, nmp->nm_nam, (struct mbuf *)0); if (error) { if (NFSIGNORE_SOERROR(nmp->nm_soflags, error)) so->so_error = 0; } else { /* * Iff first send, start timing * else turn timing off, backoff timer * and divide congestion window by 2. */ if (rep->r_flags & R_SENT) { rep->r_flags &= ~R_TIMING; if (++rep->r_rexmit > NFS_MAXREXMIT) rep->r_rexmit = NFS_MAXREXMIT; nmp->nm_cwnd >>= 1; if (nmp->nm_cwnd < NFS_CWNDSCALE) nmp->nm_cwnd = NFS_CWNDSCALE; nfsstats.rpcretries++; } else { rep->r_flags |= R_SENT; nmp->nm_sent += NFS_CWNDSCALE; } rep->r_rtt = 0; } } } #ifdef NFSSERVER /* * Scan the write gathering queues for writes that need to be * completed now. */ getmicrotime(&tv); cur_usec = (u_quad_t)tv.tv_sec * 1000000 + (u_quad_t)tv.tv_usec; TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { if (LIST_FIRST(&slp->ns_tq) && LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec) nfsrv_wakenfsd(slp); } #endif /* NFSSERVER */ splx(s); timeout_add(to, nfs_ticks); } /* * Test for a termination condition pending on the process. * This is used for NFSMNT_INT mounts. */ int nfs_sigintr(nmp, rep, p) struct nfsmount *nmp; struct nfsreq *rep; struct proc *p; { if (rep && (rep->r_flags & R_SOFTTERM)) return (EINTR); if (!(nmp->nm_flag & NFSMNT_INT)) return (0); if (p && p->p_siglist && (((p->p_siglist & ~p->p_sigmask) & ~p->p_sigignore) & NFSINT_SIGMASK)) return (EINTR); return (0); } /* * Lock a socket against others. * Necessary for STREAM sockets to ensure you get an entire rpc request/reply * and also to avoid race conditions between the processes with nfs requests * in progress when a reconnect is necessary. */ int nfs_sndlock(flagp, rep) int *flagp; struct nfsreq *rep; { struct proc *p; int slpflag = 0, slptimeo = 0; if (rep) { p = rep->r_procp; if (rep->r_nmp->nm_flag & NFSMNT_INT) slpflag = PCATCH; } else p = (struct proc *)0; while (*flagp & NFSMNT_SNDLOCK) { if (rep && nfs_sigintr(rep->r_nmp, rep, p)) return (EINTR); *flagp |= NFSMNT_WANTSND; (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsndlck", slptimeo); if (slpflag == PCATCH) { slpflag = 0; slptimeo = 2 * hz; } } *flagp |= NFSMNT_SNDLOCK; return (0); } /* * Unlock the stream socket for others. */ void nfs_sndunlock(flagp) int *flagp; { if ((*flagp & NFSMNT_SNDLOCK) == 0) panic("nfs sndunlock"); *flagp &= ~NFSMNT_SNDLOCK; if (*flagp & NFSMNT_WANTSND) { *flagp &= ~NFSMNT_WANTSND; wakeup((caddr_t)flagp); } } int nfs_rcvlock(rep) struct nfsreq *rep; { int *flagp = &rep->r_nmp->nm_flag; int slpflag, slptimeo = 0; if (*flagp & NFSMNT_INT) slpflag = PCATCH; else slpflag = 0; while (*flagp & NFSMNT_RCVLOCK) { if (nfs_sigintr(rep->r_nmp, rep, rep->r_procp)) return (EINTR); *flagp |= NFSMNT_WANTRCV; (void) tsleep((caddr_t)flagp, slpflag | (PZERO - 1), "nfsrcvlk", slptimeo); if (slpflag == PCATCH) { slpflag = 0; slptimeo = 2 * hz; } } *flagp |= NFSMNT_RCVLOCK; return (0); } /* * Unlock the stream socket for others. */ void nfs_rcvunlock(flagp) int *flagp; { if ((*flagp & NFSMNT_RCVLOCK) == 0) panic("nfs rcvunlock"); *flagp &= ~NFSMNT_RCVLOCK; if (*flagp & NFSMNT_WANTRCV) { *flagp &= ~NFSMNT_WANTRCV; wakeup((caddr_t)flagp); } } /* * Auxiliary routine to align the length of mbuf copies made with m_copyback(). */ void nfs_realign_fixup(struct mbuf *m, struct mbuf *n, unsigned int *off) { size_t padding; /* * The maximum number of bytes that m_copyback() places in a mbuf is * always an aligned quantity, so realign happens at the chain's tail. */ while (n->m_next != NULL) n = n->m_next; /* * Pad from the next elements in the source chain. Loop until the * destination chain is aligned, or the end of the source is reached. */ do { m = m->m_next; if (m == NULL) return; padding = min(ALIGN(n->m_len) - n->m_len, m->m_len); if (padding > M_TRAILINGSPACE(n)) panic("nfs_realign_fixup: no memory to pad to"); bcopy(mtod(m, void *), mtod(n, char *) + n->m_len, padding); n->m_len += padding; m_adj(m, padding); *off += padding; } while (!ALIGNED_POINTER(n->m_len, void *)); } /* * The NFS RPC parsing code uses the data address and the length of mbuf * structures to calculate on-memory addresses. This function makes sure these * parameters are correctly aligned. */ void nfs_realign(struct mbuf **pm, int hsiz) { struct mbuf *m; struct mbuf *n = NULL; unsigned int off = 0; ++nfs_realign_test; while ((m = *pm) != NULL) { if (!ALIGNED_POINTER(m->m_data, void *) || !ALIGNED_POINTER(m->m_len, void *)) { MGET(n, M_WAIT, MT_DATA); if (ALIGN(m->m_len) >= MINCLSIZE) { MCLGET(n, M_WAIT); } n->m_len = 0; break; } pm = &m->m_next; } /* * If n is non-NULL, loop on m copying data, then replace the * portion of the chain that had to be realigned. */ if (n != NULL) { ++nfs_realign_count; while (m) { m_copyback(n, off, m->m_len, mtod(m, caddr_t)); /* * If an unaligned amount of memory was copied, fix up * the last mbuf created by m_copyback(). */ if (!ALIGNED_POINTER(m->m_len, void *)) nfs_realign_fixup(m, n, &off); off += m->m_len; m = m->m_next; } m_freem(*pm); *pm = n; } } /* * Parse an RPC request * - verify it * - fill in the cred struct. */ int nfs_getreq(nd, nfsd, has_header) struct nfsrv_descript *nd; struct nfsd *nfsd; int has_header; { int len, i; u_int32_t *tl; int32_t t1; struct uio uio; struct iovec iov; caddr_t dpos, cp2, cp; u_int32_t nfsvers, auth_type; uid_t nickuid; int error = 0, ticklen; struct mbuf *mrep, *md; struct nfsuid *nuidp; struct timeval tvin, tvout; mrep = nd->nd_mrep; md = nd->nd_md; dpos = nd->nd_dpos; if (has_header) { nfsm_dissect(tl, u_int32_t *, 10 * NFSX_UNSIGNED); nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++); if (*tl++ != rpc_call) { m_freem(mrep); return (EBADRPC); } } else nfsm_dissect(tl, u_int32_t *, 8 * NFSX_UNSIGNED); nd->nd_repstat = 0; nd->nd_flag = 0; if (*tl++ != rpc_vers) { nd->nd_repstat = ERPCMISMATCH; nd->nd_procnum = NFSPROC_NOOP; return (0); } if (*tl != nfs_prog) { nd->nd_repstat = EPROGUNAVAIL; nd->nd_procnum = NFSPROC_NOOP; return (0); } tl++; nfsvers = fxdr_unsigned(u_int32_t, *tl++); if (nfsvers != NFS_VER2 && nfsvers != NFS_VER3) { nd->nd_repstat = EPROGMISMATCH; nd->nd_procnum = NFSPROC_NOOP; return (0); } if (nfsvers == NFS_VER3) nd->nd_flag = ND_NFSV3; nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++); if (nd->nd_procnum == NFSPROC_NULL) return (0); if (nd->nd_procnum >= NFS_NPROCS || (nd->nd_procnum > NFSPROC_COMMIT) || (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { nd->nd_repstat = EPROCUNAVAIL; nd->nd_procnum = NFSPROC_NOOP; return (0); } if ((nd->nd_flag & ND_NFSV3) == 0) nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; auth_type = *tl++; len = fxdr_unsigned(int, *tl++); if (len < 0 || len > RPCAUTH_MAXSIZ) { m_freem(mrep); return (EBADRPC); } nd->nd_flag &= ~ND_KERBAUTH; /* * Handle auth_unix or auth_kerb. */ if (auth_type == rpc_auth_unix) { len = fxdr_unsigned(int, *++tl); if (len < 0 || len > NFS_MAXNAMLEN) { m_freem(mrep); return (EBADRPC); } nfsm_adv(nfsm_rndup(len)); nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); bzero((caddr_t)&nd->nd_cr, sizeof (struct ucred)); nd->nd_cr.cr_ref = 1; nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++); nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++); len = fxdr_unsigned(int, *tl); if (len < 0 || len > RPCAUTH_UNIXGIDS) { m_freem(mrep); return (EBADRPC); } nfsm_dissect(tl, u_int32_t *, (len + 2) * NFSX_UNSIGNED); for (i = 0; i < len; i++) if (i < NGROUPS) nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); else tl++; nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len; if (nd->nd_cr.cr_ngroups > 1) nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups); len = fxdr_unsigned(int, *++tl); if (len < 0 || len > RPCAUTH_MAXSIZ) { m_freem(mrep); return (EBADRPC); } if (len > 0) nfsm_adv(nfsm_rndup(len)); } else if (auth_type == rpc_auth_kerb) { switch (fxdr_unsigned(int, *tl++)) { case RPCAKN_FULLNAME: ticklen = fxdr_unsigned(int, *tl); *((u_int32_t *)nfsd->nfsd_authstr) = *tl; uio.uio_resid = nfsm_rndup(ticklen) + NFSX_UNSIGNED; nfsd->nfsd_authlen = uio.uio_resid + NFSX_UNSIGNED; if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) { m_freem(mrep); return (EBADRPC); } uio.uio_offset = 0; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_segflg = UIO_SYSSPACE; iov.iov_base = (caddr_t)&nfsd->nfsd_authstr[4]; iov.iov_len = RPCAUTH_MAXSIZ - 4; nfsm_mtouio(&uio, uio.uio_resid); nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); if (*tl++ != rpc_auth_kerb || fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { printf("Bad kerb verifier\n"); nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); nd->nd_procnum = NFSPROC_NOOP; return (0); } nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); tl = (u_int32_t *)cp; if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { printf("Not fullname kerb verifier\n"); nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); nd->nd_procnum = NFSPROC_NOOP; return (0); } cp += NFSX_UNSIGNED; bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED); nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; nd->nd_flag |= ND_KERBFULL; nfsd->nfsd_flag |= NFSD_NEEDAUTH; break; case RPCAKN_NICKNAME: if (len != 2 * NFSX_UNSIGNED) { printf("Kerb nickname short\n"); nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); nd->nd_procnum = NFSPROC_NOOP; return (0); } nickuid = fxdr_unsigned(uid_t, *tl); nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED); if (*tl++ != rpc_auth_kerb || fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { printf("Kerb nick verifier bad\n"); nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); nd->nd_procnum = NFSPROC_NOOP; return (0); } nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED); tvin.tv_sec = *tl++; tvin.tv_usec = *tl; LIST_FOREACH(nuidp, NUIDHASH(nfsd->nfsd_slp, nickuid), nu_hash) { if (nuidp->nu_cr.cr_uid == nickuid && (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), &nuidp->nu_haddr, nd->nd_nam2))) break; } if (!nuidp) { nd->nd_repstat = (NFSERR_AUTHERR|AUTH_REJECTCRED); nd->nd_procnum = NFSPROC_NOOP; return (0); } tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); if (nuidp->nu_expire < time_second || nuidp->nu_timestamp.tv_sec > tvout.tv_sec || (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { nuidp->nu_expire = 0; nd->nd_repstat = (NFSERR_AUTHERR|AUTH_REJECTVERF); nd->nd_procnum = NFSPROC_NOOP; return (0); } nfsrv_setcred(&nuidp->nu_cr, &nd->nd_cr); nd->nd_flag |= ND_KERBNICK; }; } else { nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); nd->nd_procnum = NFSPROC_NOOP; return (0); } nd->nd_md = md; nd->nd_dpos = dpos; return (0); nfsmout: return (error); } int nfs_msg(p, server, msg) struct proc *p; char *server, *msg; { tpr_t tpr; if (p) tpr = tprintf_open(p); else tpr = NULL; tprintf(tpr, "nfs server %s: %s\n", server, msg); tprintf_close(tpr); return (0); } #ifdef NFSSERVER int (*nfsrv3_procs[NFS_NPROCS])(struct nfsrv_descript *, struct nfssvc_sock *, struct proc *, struct mbuf **) = { nfsrv_null, nfsrv_getattr, nfsrv_setattr, nfsrv_lookup, nfsrv3_access, nfsrv_readlink, nfsrv_read, nfsrv_write, nfsrv_create, nfsrv_mkdir, nfsrv_symlink, nfsrv_mknod, nfsrv_remove, nfsrv_rmdir, nfsrv_rename, nfsrv_link, nfsrv_readdir, nfsrv_readdirplus, nfsrv_statfs, nfsrv_fsinfo, nfsrv_pathconf, nfsrv_commit, nfsrv_noop, nfsrv_noop, nfsrv_noop, nfsrv_noop }; /* * Socket upcall routine for the nfsd sockets. * The caddr_t arg is a pointer to the "struct nfssvc_sock". * Essentially do as much as possible non-blocking, else punt and it will * be called with M_WAIT from an nfsd. */ void nfsrv_rcv(so, arg, waitflag) struct socket *so; caddr_t arg; int waitflag; { struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; struct mbuf *m; struct mbuf *mp, *nam; struct uio auio; int flags, error; if ((slp->ns_flag & SLP_VALID) == 0) return; #ifdef notdef /* * Define this to test for nfsds handling this under heavy load. */ if (waitflag == M_DONTWAIT) { slp->ns_flag |= SLP_NEEDQ; goto dorecs; } #endif auio.uio_procp = NULL; if (so->so_type == SOCK_STREAM) { /* * If there are already records on the queue, defer soreceive() * to an nfsd so that there is feedback to the TCP layer that * the nfs servers are heavily loaded. */ if (slp->ns_rec && waitflag == M_DONTWAIT) { slp->ns_flag |= SLP_NEEDQ; goto dorecs; } /* * Do soreceive(). */ auio.uio_resid = 1000000000; flags = MSG_DONTWAIT; error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); if (error || mp == (struct mbuf *)0) { if (error == EWOULDBLOCK) slp->ns_flag |= SLP_NEEDQ; else slp->ns_flag |= SLP_DISCONN; goto dorecs; } m = mp; if (slp->ns_rawend) { slp->ns_rawend->m_next = m; slp->ns_cc += 1000000000 - auio.uio_resid; } else { slp->ns_raw = m; slp->ns_cc = 1000000000 - auio.uio_resid; } while (m->m_next) m = m->m_next; slp->ns_rawend = m; /* * Now try and parse record(s) out of the raw stream data. */ error = nfsrv_getstream(slp, waitflag); if (error) { if (error == EPERM) slp->ns_flag |= SLP_DISCONN; else slp->ns_flag |= SLP_NEEDQ; } } else { do { auio.uio_resid = 1000000000; flags = MSG_DONTWAIT; error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags); if (mp) { if (nam) { m = nam; m->m_next = mp; } else m = mp; if (slp->ns_recend) slp->ns_recend->m_nextpkt = m; else slp->ns_rec = m; slp->ns_recend = m; m->m_nextpkt = (struct mbuf *)0; } if (error) { if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && error != EWOULDBLOCK) { slp->ns_flag |= SLP_DISCONN; goto dorecs; } } } while (mp); } /* * Now try and process the request records, non-blocking. */ dorecs: if (waitflag == M_DONTWAIT && (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) nfsrv_wakenfsd(slp); } /* * Try and extract an RPC request from the mbuf data list received on a * stream socket. The "waitflag" argument indicates whether or not it * can sleep. */ int nfsrv_getstream(slp, waitflag) struct nfssvc_sock *slp; int waitflag; { struct mbuf *m, **mpp; char *cp1, *cp2; int len; struct mbuf *om, *m2, *recm; u_int32_t recmark; if (slp->ns_flag & SLP_GETSTREAM) panic("nfs getstream"); slp->ns_flag |= SLP_GETSTREAM; for (;;) { if (slp->ns_reclen == 0) { if (slp->ns_cc < NFSX_UNSIGNED) { slp->ns_flag &= ~SLP_GETSTREAM; return (0); } m = slp->ns_raw; if (m->m_len >= NFSX_UNSIGNED) { bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED); m->m_data += NFSX_UNSIGNED; m->m_len -= NFSX_UNSIGNED; } else { cp1 = (caddr_t)&recmark; cp2 = mtod(m, caddr_t); while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) { while (m->m_len == 0) { m = m->m_next; cp2 = mtod(m, caddr_t); } *cp1++ = *cp2++; m->m_data++; m->m_len--; } } slp->ns_cc -= NFSX_UNSIGNED; recmark = ntohl(recmark); slp->ns_reclen = recmark & ~0x80000000; if (recmark & 0x80000000) slp->ns_flag |= SLP_LASTFRAG; else slp->ns_flag &= ~SLP_LASTFRAG; if (slp->ns_reclen > NFS_MAXPACKET) { slp->ns_flag &= ~SLP_GETSTREAM; return (EPERM); } } /* * Now get the record part. */ recm = NULL; if (slp->ns_cc == slp->ns_reclen) { recm = slp->ns_raw; slp->ns_raw = slp->ns_rawend = (struct mbuf *)0; slp->ns_cc = slp->ns_reclen = 0; } else if (slp->ns_cc > slp->ns_reclen) { len = 0; m = slp->ns_raw; om = (struct mbuf *)0; while (len < slp->ns_reclen) { if ((len + m->m_len) > slp->ns_reclen) { m2 = m_copym(m, 0, slp->ns_reclen - len, waitflag); if (m2) { if (om) { om->m_next = m2; recm = slp->ns_raw; } else recm = m2; m->m_data += slp->ns_reclen - len; m->m_len -= slp->ns_reclen - len; len = slp->ns_reclen; } else { slp->ns_flag &= ~SLP_GETSTREAM; return (EWOULDBLOCK); } } else if ((len + m->m_len) == slp->ns_reclen) { om = m; len += m->m_len; m = m->m_next; recm = slp->ns_raw; om->m_next = (struct mbuf *)0; } else { om = m; len += m->m_len; m = m->m_next; } } slp->ns_raw = m; slp->ns_cc -= len; slp->ns_reclen = 0; } else { slp->ns_flag &= ~SLP_GETSTREAM; return (0); } /* * Accumulate the fragments into a record. */ mpp = &slp->ns_frag; while (*mpp) mpp = &((*mpp)->m_next); *mpp = recm; if (slp->ns_flag & SLP_LASTFRAG) { if (slp->ns_recend) slp->ns_recend->m_nextpkt = slp->ns_frag; else slp->ns_rec = slp->ns_frag; slp->ns_recend = slp->ns_frag; slp->ns_frag = (struct mbuf *)0; } } } /* * Parse an RPC header. */ int nfsrv_dorec(slp, nfsd, ndp) struct nfssvc_sock *slp; struct nfsd *nfsd; struct nfsrv_descript **ndp; { struct mbuf *m, *nam; struct nfsrv_descript *nd; int error; *ndp = NULL; if ((slp->ns_flag & SLP_VALID) == 0 || (m = slp->ns_rec) == (struct mbuf *)0) return (ENOBUFS); slp->ns_rec = m->m_nextpkt; if (slp->ns_rec) m->m_nextpkt = (struct mbuf *)0; else slp->ns_recend = (struct mbuf *)0; if (m->m_type == MT_SONAME) { nam = m; m = m->m_next; nam->m_next = NULL; } else nam = NULL; nd = malloc(sizeof(struct nfsrv_descript), M_NFSRVDESC, M_WAITOK); nfs_realign(&m, 10 * NFSX_UNSIGNED); nd->nd_md = nd->nd_mrep = m; nd->nd_nam2 = nam; nd->nd_dpos = mtod(m, caddr_t); error = nfs_getreq(nd, nfsd, TRUE); if (error) { m_freem(nam); free(nd, M_NFSRVDESC); return (error); } *ndp = nd; nfsd->nfsd_nd = nd; return (0); } /* * Search for a sleeping nfsd and wake it up. * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the * running nfsds will go look for the work in the nfssvc_sock list. */ void nfsrv_wakenfsd(slp) struct nfssvc_sock *slp; { struct nfsd *nd; if ((slp->ns_flag & SLP_VALID) == 0) return; for (nd = TAILQ_FIRST(&nfsd_head); nd != NULL; nd = TAILQ_NEXT(nd, nfsd_chain)) { if (nd->nfsd_flag & NFSD_WAITING) { nd->nfsd_flag &= ~NFSD_WAITING; if (nd->nfsd_slp) panic("nfsd wakeup"); slp->ns_sref++; nd->nfsd_slp = slp; wakeup((caddr_t)nd); return; } } slp->ns_flag |= SLP_DOREC; nfsd_head_flag |= NFSD_CHECKSLP; } #endif /* NFSSERVER */